Feat/mig: Multi Instance GPU (MIG)

[venkata22a]

What type of PR is this?

• Bug fix
• New feature
• Documentation
• Refactoring
• CI/Build

Description

Support for MIG Metrics across the environments.

Related Issue

#26

Checklist

• Tests added/updated
• Documentation updated (if applicable)
• make test passes
• make lint passes
• Commits are signed off (git commit -s)

[pmady]

nice work on the MIG support, solid feature. couple things to address before merge:

1. MIGUUIDs() is copy-pasted identically in both pkg/slurm/slurm.go and pkg/flux/flux.go — same loop, same strings.HasPrefix(p, "MIG-") check. can you pull that into a shared helper? maybe something like gpu.ParseMIGUUIDs(csv string) []string or a util in pkg/env.

2. the new MIG fields on Metrics struct (IsMIGInstance, ParentIndex, MigProfile) don't have json tags. the rest of the struct uses explicit json tags — these should too for consistency and to match the csv output field names. something like:

IsMIGInstance bool   `json:"is_mig_instance"`
ParentIndex   int    `json:"parent_index"`
MigProfile    string `json:"mig_profile,omitempty"`

3. CollectByUUID hardcodes instanceIdx: 0 when the real index isn't known from a UUID lookup. worth adding a comment explaining this, or see if GetIndex() works on MIG device handles to get the actual instance index.

4. this PR has a merge commit from main that includes the cross-env code from #59. should we merge #59 first and rebase this on top? cleaner git history.

overall looks good — the physical metrics sharing approach is the right call and the test coverage on aggregation across MIG instances is solid.